Deep-learning-assisted wedge fiber optic surface plasmon resonance sensor for accuracy detection of trace mercury ions

The increasing levels of heavy metal pollution pose significant health and environmental challenges, particularly the existence of mercury ions ( Hg 2+ ). Even trace concentrations of Hg 2+ can cause serious harm to the human body, making highly sensitive and accurate detection of tiny concentrations of Hg 2+ particularly important. This study demonstrated a novel deep-learning-assisted fiber optic surface plasmon resonance (SPR) sensor for Hg 2+ detection. The fiber optic sensor utilized a wedge probe with an incidence angle of 73° and self-assembly of gold nanospheres (AuNPs) on the surface of the sensing region by DL-dithiothreitol (DTT). Experimental results demonstrated that the sensor exhibited a refractive index (RI) sensitivity of up to 30,400 nm/RIU, which is significantly higher than many previously reported values, and enabled the detection of Hg 2+ at concentrations as low as 1 pmol/L. Furthermore, the fiber optic sensor was tested in 12 (1 pmol/L to 500 nmol/L) solutions with different Hg 2+ concentrations, and the corresponding SPR spectra were recorded. A residual neural network (ResNet)-based deep learning model was developed to classify and recognize the SPR spectra of 12 different Hg 2+ concentrations with a training accuracy of 99.76%. In a blind test on 120 samples, the ResNet model had a prediction accuracy of 99.17%. Therefore, the combination of the sensor and deep learning algorithm proposed in this study holds significant potential for a wide range of practical applications.